The wage gap discussion in the US has long struck me as being a little strange. One side says that women make 20 percent less than men. Then the other side says that, once you control for the different labor market characteristics of men and women, the gap is much smaller. Then the first side comes back and says that the fact that men and women have different labor market characteristics is itself gendered and that this gendered sorting is still a problem.

The latter point has been the sophisticated argument about the gender wage gap for some time, an argument that seems quite persuasive to me. Unequal pay for identical work is not the only way that a labor market can be sexist. A labor market that sorts men into higher-paying jobs and women into lower-paying jobs is still sexist, just in a different way.

But what’s odd about this sophisticated argument is that its advocates all seem to be using the wrong statistics. The commonly cited National Women’s Law Center (NWLC) statistic that says women make 80 percent of what men make only compares the median earnings of full-time workers. So despite making a very compelling case against controlling for labor market characteristics, advocates end up using one of those controls for their own headline statistic. Just as it ignores gendered sorting to control for job type, hours worked, and experience, it also ignores gendered sorting to control for full-time work.

The correct statistic for those partial to the sophisticated argument should be a completely uncontrolled comparison of the median earnings of all men and women who are in their prime working years (ages twenty-five to fifty-four), including those who work less than full time and those who do not do any paid work. Unlike the NWLC statistic, this comparison reflects the gendered sorting that results in women being more likely to have part-time work and more likely to be out of the labor market altogether.

So what is the result of such a calculation? Using the latest census data, I find that the median woman between the ages of twenty-five and fifty-four earned $25,000 in 2016. For men, it was $41,000. This means that women earn 39 percent less than men, not 20 percent.

The difference between the conventional 20 percent finding and the correct 39 percent finding is, as you would expect, completely driven by gendered differences in who engages in full-time work. When I compare full-time workers (those working fifty or more weeks) between the ages of twenty-five and fifty-four, I find the median woman earning $40,000 and the median man earning $50,000. Even though this calculation is slightly different than the NWLC wage gap, it nonetheless shows women earning 20 percent less than men. When I compare all workers (those working one or more weeks) between the ages of twenty-five and fifty-four, I get women earning $35,000 and men earning $47,750, meaning a wage gap of 27 percent. And then when I compare all people in that age band, including non-workers, I get the 39 percent wage gap figure.

If you are trying to determine how much less employers pay women for the same work, then you should use a bunch of controls to ensure an apples-to-apples labor comparison. If you are trying to determine what effect gendered sorting in the labor market has on earnings, then you should use no controls, as I do here. The conventional NWLC measure that says that women make 80 cents on the dollar uses one control (full-time work) while rejecting all the other controls, which places it in a weird middle zone that does not track anything of clear significance.